M
Massimo
Guest
I've been working a cluster of two Windows 2003 Enterprise SP2 systems,
which run SQL Server and are also domain controllers for a local domain
(it's an hosted environment, so only two servers could be fitted there). The
two servers have each 18 GB of memory, but run the x86 Windows version.
The /PAE switch is enabled on those systems, in order to take advantage of
all that memory; today, we tried enabling the /3GB switch too, to give SQL
Server some extra memory... but then the systems started behaving
erratically, taking hours (literally!) to start up, even more time to log
on, showing plenty of errors in the event log about almost everything
crashing randomly and not letting you starting any program when you finally
got to the desktop; a typical memory error condition. This happened on the
passive cluster node, on which we tried the new settings, so it was not
related to SQL Server at all.
Everything went back to normal when the /3GB switch was removed.
Why is this happening? I've found some articles stating that the /3GB switch
can cause problems with memory-hungry device drivers, such as 3D video
cards, but those are server systems and most definitely don't have that kind
of hardware; I've also read that due to page table problems, you can't use
more than 16 GB of memory when both /PAE and /3GB are enabled; so I tried
removing 2 GB of memory from them, but the problem persisted: when /3GB is
enabled, the systems gets completely unstable. When only /PAE is enabled,
everything works fine, but (obviously) programs can't use more than 2 GB
each.
I've enable both /3GB and /PAE on other systems without having any problem,
but they had less physical memory, and they were also not domain
controllers. Could this be a problem of kernel memory exhaustion due to
excessive page table usage and/or kernel memory requirements of the domain
controller role?
Can someone please help?
Thanks
Massimo
which run SQL Server and are also domain controllers for a local domain
(it's an hosted environment, so only two servers could be fitted there). The
two servers have each 18 GB of memory, but run the x86 Windows version.
The /PAE switch is enabled on those systems, in order to take advantage of
all that memory; today, we tried enabling the /3GB switch too, to give SQL
Server some extra memory... but then the systems started behaving
erratically, taking hours (literally!) to start up, even more time to log
on, showing plenty of errors in the event log about almost everything
crashing randomly and not letting you starting any program when you finally
got to the desktop; a typical memory error condition. This happened on the
passive cluster node, on which we tried the new settings, so it was not
related to SQL Server at all.
Everything went back to normal when the /3GB switch was removed.
Why is this happening? I've found some articles stating that the /3GB switch
can cause problems with memory-hungry device drivers, such as 3D video
cards, but those are server systems and most definitely don't have that kind
of hardware; I've also read that due to page table problems, you can't use
more than 16 GB of memory when both /PAE and /3GB are enabled; so I tried
removing 2 GB of memory from them, but the problem persisted: when /3GB is
enabled, the systems gets completely unstable. When only /PAE is enabled,
everything works fine, but (obviously) programs can't use more than 2 GB
each.
I've enable both /3GB and /PAE on other systems without having any problem,
but they had less physical memory, and they were also not domain
controllers. Could this be a problem of kernel memory exhaustion due to
excessive page table usage and/or kernel memory requirements of the domain
controller role?
Can someone please help?
Thanks
Massimo